The course is under construction, with new materials appearing regularly.
Subscribe for updates and make sure you don’t miss anything: Stay updated
This is an open course on using and deploying LLM APIs and open-source LLMs.
Materials for each topic can be found in the ./topic* folders. See README.md for further details and instructions.
The course includes a capstone project: NPC (Non-Playable Character) Factory Development. In this project, you'll build a service based on either an LLM API or an open-source LLM to provide NPC interactions for users. You'll go through the entire process, from personalizing a chat bot to making a cost-and-quality decision between using an API or hosting an LLM on your own servers.
Any technical issues, ideas, bugs in course materials, contribution ideas - add an issue.
Basics of LLM and Multimodal LLM API usage and prompt strategies, typical problems arising with LLMs, creativity-vs-reproducibility control.
Project: Creating a chatbot and deploying it in a cloud.
LLM Workflows and beyond: from Chaining to AI Agents. LLM Reasoning.
Project: Planning and memory summarization; Automating evaluation.
RAG and its technicalities; vector stores, databases in production. RAG evaluation.
Project: Adding RAG to the NPC Factory service.
Working with open source LLMs and practical LLM inference in production. Computational and memory bottlenecks of LLM inference.
Project: Deploying a chat service based on a self-served LLM. Serving text encoders and rerankers. Making a cost-to-value choice between API and self-served LLMs.
Optimizing LLM inference, quantization and beyond. Production monitoring and observability.
Project: Optimizing open source LLM inference. Establishing monitoring with Evidently AI, Prometeus and Grafana.
Fine tuning of LLMs and embeddings. Parameter-Efficient Fine Tuning and LoRA. RLHF and DPO
Project: Making your characters even more alive through fine tuning